Developmental language disorder – a comprehensive study of more than 46,000 individuals

Developmental language disorder (DLD) is characterized by enduring low language abilities with a significant functional impact, in the absence of biomedical conditions where language impairment is part of a complex of impairments. There is a lack of awareness of DLD even among healthcare professionals. Here we estimated the prevalence of DLD and its connections to reading and learning difficulties and physical and mental health in the Danish Blood Donor Study (N=46,547), where DLD-related information is based on questionnaires (self-report). We compared the questionnaire-derived DLD status with the relevant language-related diagnoses from hospital registers. We investigated the genetic architecture of DLD in a subset of the cohort (N=18,380). DLD was significantly associated with reading and learning difficulties and poorer mental and physical health. DLD prevalence was 3.36%–3.70% based on questionnaires, compared with 0.04% in hospital registers. Our genetic analyses identified one genome-wide significant locus, but not a significant heritability estimate. Our study shows that DLD has health-related implications that may last into adulthood, and that DLD may be undiagnosed in general healthcare. Furthermore, DLD is likely genetically heterogeneous. Our results emphasize the need to raise awareness of DLD and consider criteria for molecular studies of DLD to reduce case heterogeneity.


INTRODUCTION
Developmental language disorder (DLD) is defined as enduring substantially low language abilities which significantly affect daily functioning, in the absence of biomedical conditions (e.g. sensory-neural hearing loss, intellectual impairment or autism spectrum disorder) where language impairment is only part of a complex of impairments (Bishop, Snowling, Thompson, & Greenhalgh, 2017). DLD is associated with negative consequences, especially on social and emotional development and learning, and the condition and its implications often persist into adulthood (Elbro, Dalby, & Maarbjerg, 2011). To reach a consensus on both the concept and the label, two Delphi studies by the DLD consortium (Bishop, Snowling, Thompson, & Greenhalgh, 2016;Bishop, et al., 2017) were conducted, resulting in the term DLD. The previous, common label for this condition, specific language impairment (SLI), required a discrepancy between language ability and non-verbal IQ (Bishop, 2006). However, there was no consensus on the extent of this discrepancy, or on the cut-off threshold for language impairment. SLI research has mainly included participants with language scores of more than 1.25 or 1.5 standard deviations (SD) below the population mean, with inclusion thresholds for non-verbal IQ varying between 70 and 85. The first SLI Consortium genome-wide linkage study included children with SLI who had language scores ≥1.5 SD below the population mean and non-verbal IQ ≥80 (The SLI Consortium, 2002). A study from The UK reported a prevalence of ~4.80% (Norbury et al., 2016). The criteria for DLD (language disorder of unknown origin) in that study were scores ≥1.5 SD below the population mean for at least two of five language composite scores and nonverbal IQ >85. With relaxed inclusion criteria for nonverbal IQ (>70) the estimated prevalence was 7.58%. When Norbury et al. used the Tomblin et al. criteria, which have been widely used in the SLI research, they obtained a prevalence of 7.74%, which is similar to the SLI prevalence reported by the latter (Tomblin et al., 1997), namely 7.4%. However, when ICD-10 criteria with a requirement for significant discrepancy between language ability and nonverbal IQ were employed, the prevalence estimate dropped to 1.07% (Norbury, et al., 2016). The lack of consensus regarding diagnostic criteria and terminology has hampered both research efforts and access to support services (Reilly, Bishop, & Tomblin, 2014). In Denmark, there are no national guidelines for DLD, and there is no obligation to provide specialized services to children with speech, language and communication disorders (de López & Knudsen, 2019). Speech and language therapy services do not require a clinical diagnosis (they are provided to children based on identification of their difficulties). These services are largely managed publicly at the municipal level, resulting in considerable variation.
Moreover, increased inclusive education over the last decades have resulted in a decrease in special education for children with DLD and an increase in consultation-based services for teachers and parents (de López & Knudsen, 2019).

Developmental language disorderhealth outcomes and comorbidities
Individuals with DLD have an increased risk of poor health-related quality of life (Lyons, 2021), including mental health problems (Botting, Toseeb, Pickles, Durkin, & Conti-Ramsden, 2016;Eadie et al., 2018;Y. C. Lee et al., 2020;Lyons, 2021). A British longterm follow-up study reported higher prevalences of anxiety and depression in adolescents and adults with DLD than in controls (Botting, et al., 2016). Moreover, DLD is associated with an increased risk of reading impairments, such as poor reading comprehension and poor reading decoding skills (Botting, 2020;Kalnak & Sahlen, 2022), 6 with long-term effects on learning and education (Conti-Ramsden, Durkin, Toseeb, Botting, & Pickles, 2018) and employment (Elbro, et al., 2011). However, there is considerable heterogeneity among individuals with DLD; not all have poor mental health or negative outcomes in learning and education (Le et al., 2021).

Heritability estimates and known genetic risk loci of developmental language disorders
Heritability is defined as the proportion of the variation in a trait that is due to genetic variation across individuals in a given population (Mayhew & Meyre, 2017). Evidence for the heritability of diseases can be obtained from twin data by examining the monozygotic (MZ) and dizygotic (DZ) twin concordance rates; a higher MZ concordance rate suggests a genetic component. For (spoken) language disorders, a meta-analysis (based also on studies employing SLI criteria) obtained an overall MZ concordance rate of 83.6% and an overall DZ concordance rate of 50.2%, which were significantly different (Stromswold, 2001). Earlier work using twins (with at least one in each pair meeting criteria for SLI) and a DeFries-Fulker analysis (DeFries & Fulker, 1988) showed high heritabilities for some language traits (Bishop, North, & Donlan, 1995). Some studies have estimated the heritabilities of language-related traits (Bishop, et al., 1995;Rice, Taylor, Zubrick, Hoffman, & Earnest, 2020), but, to our knowledge, no studies of language impairment have examined the heritability derived from genetic associations from genome-wide association studies (GWAS), namely, the narrow-sense singlenucleotide polymorphism (SNP) heritability, or h 2 SNP , for DLD. Narrow-sense heritability (h 2 ) is defined as the proportion of the variation in a trait that is due to additive genetic variation across individuals in a given population (Mayhew & Meyre, 2017). It follows that h 2 SNP is the narrow-sense heritability that is attributed to the combined effects of SNPs (included in a study), which are genetic loci of common variation. Recently, a polygenic risk score (PRS, representing an individual"s genetic predisposition to a trait or a disease) for SLI was shown to be associated with SLI in an independent sample (Nudel, Appadurai, Buil, Nordentoft, & Werge, 2021;Nudel et al., 2020). The PRS is limited by h 2 , and it approaches it as GWASs have bigger samples and more markers (Wray et al., 2021). But a PRS used in this way may be further limited by several factors, including the power of the discovery GWAS, ascertainment differences between the discovery and target studies and the overlap in the marker datasets between the studies. Therefore, we sought to estimate h 2 SNP for DLD directly. Several specific loci for language disorders have been identified. The first genome-wide linkage studies for SLI identified genomic loci linked to nonword repetition ability (NWR) and expressive language in children with SLI on chromosomes 16 and 19, respectively (The SLI Consortium, 2002, 2004, and a region on chromosome 13 was linked to reading ability in children with SLI (Bartlett et al., 2002). Further association analyses highlighted two genes on chromosome 16 associated with NWR, namely CMIP and ATP2C2 (Newbury et al., 2009). Other highlighted genes include NOP9, identified in a GWAS for SLI modeling paternal parent-of-origin effects (Nudel et al., 2014b), CNTNAP2, also associated with NWR (Vernes et al., 2008), NFXL1, identified in an exome-sequencing study of an isolated population with a high prevalence of SLI (Villanueva et al., 2015), and several other genes included in studies of immune-related genes and whole-exome sequencing studies (Andres, Earnest, Zhong, Rice, & Raza, 2021;Chen et al., 2017;Nudel et al., 2014a).
Lastly, copy number variants have also been implicated in SLI and DLD (Kalnak et al.,8 2018; Simpson et al., 2015). When examining genes involved in broader language-related phenotypes, the list grows much bigger, as reviewed by others (Guerra & Cacabelos, 2019).

Aims of the study
The general aims of this study were: 1. To elucidate the epidemiological, societal and health-related aspects of DLD within a generally healthy adult population cohort, the Danish Blood Donor Study (DBDS).
2. To estimate the genetic contribution to DLD, including the identification of specific genetic risk variants in the same cohort.
These aims lead to the following research questions: 1a) What is the prevalence of DLD in a population sample of healthy adults and how does it compare with the incidence of formal diagnoses of language disorder as recorded in the health registers? 1b) What are the trends in special education services for participants with DLD and how have they evolved over the past decades? 1c) What is the prevalence of comorbid DLD and reading and learning difficulties? 1d) Are there differences in general mental and/or physical health measures between DLD cases and controls? 2a) What is the heritability of DLD, especially in the context of prior estimates of the heritability of SLI? 2b) Are there any specific genetic risk variants for DLD, and are there overlaps with previously reported loci for SLI?
Information about participants" diagnoses of neurodevelopmental diagnosis, including language disorders (in the form of ICD-8 and ICD-10 codes), was obtained from the Danish National Patient Register (Andersen, Madsen, Jorgensen, Mellemkjoer, & Olsen, 1999) and the Danish Psychiatric Central Research Register (Mors, Perto, & Mortensen, 2011). A subset of the sample was genotyped as part of the DBDS Genomic Consortium and subsequently used in the genetic analyses in this study (Hansen, et al., 2019). English translation). DLD cases were defined as i) having or having had a self-reported language disorder or ii) having started talking late, and, additionally, having a history of at least one of the following types of intervention: iii) speech and language therapy; iv) language group intervention; v) school language unit; vi) another form of language support. Controls were defined as i) not having or having had a self-reported language disorder, and ii) not reporting started talking late, and iii) not having indicated that others had difficulty understanding them at age 5 (see flowchart outlining the phenotype definitions in Supplementary Material). DLD cases had the following exclusionary criteria based on primary and secondary diagnoses from hospital registers for the following conditions: intellectual disability, autism spectrum disorder, hearing impairment and deafness, Down syndrome, epilepsy, Turner syndrome, Klinefelter syndrome (Supplementary Table S1).

Additional phenotypes
Questions about reading and learning difficulties were included in the DBDS questionnaire. We also composed questions about interventions for reading impairment for the purpose of this study (Supplementary Material). To investigate participants" health status, we included the physical component summary score and the mental health component summary score from the Short Form-12 (SF-12) health survey (Jenkinson et al., 1997). The SF-12 was translated into Danish and included in the DBDS questionnaire. Higher scores indicate better health, and scores range from 0 to 100, as described in detail previously (Dinh et al., 2019). Our study used non-imputed scores.

Statistical analyses
Statistical analyses were performed in R (R Core Team, 2014) v4.0.0. The tests for association between age at the submission of the questionnaire and DLD status, and sex and DLD status, were performed with the wilcox.test (U test, two-sided) and the chisq.test (χ 2 test) functions, respectively. Odds ratios (ORs) and 95% confidence intervals for the latter were calculated with the fisher.test function. Linear regressions of the mental health component summary score and the physical component summary score on DLD status were performed with the lm function and the 95% confidence intervals (CIs) were calculated with the confint function. Covariates for age and sex were included in the regression models, since they explain some of the variability in the two component scores (Steenstrup, Pedersen, Hjelmborg, Skytthe, & Kyvik, 2013). Similarly, logistic regressions of reading impairment and learning difficulties (after removing individuals with missing or uncertain answers to the relevant questions) on DLD status were performed with the glm function and the 95% confidence intervals (CIs) were calculated with the confint function. Covariates for age and sex were included in these models as well, as there are known sex differences with regards to both reading impairment and learning difficulties (Nass, 1993), and there are also age-related effects in both conditions (Holland, 2000;Wagner et al., 2020). Given the large age range in our sample and the fact that the chances of getting a neurodevelopmental diagnosis decrease with age, adding a covariate for age allowed us to better isolate the effect of DLD and also control for potential confounding.
The p-values for the regression coefficients are two-sided.

Genetic quality control for genotyping and samples
We had genotype data for a subset of the cohort, which were processed and made available as the "DBDS September 23 2019 data freeze". The samples were genotyped on the Illumina Global Screening Array 24v.1. Initial QC for this cohort is described elsewhere (Hansen, et al., 2019). Before imputation, individuals with >0.03 missing genotypes and/or sex mismatches between genetic data and register data were removed with PLINK (Purcell et al., 2007) v2.00alpha20190429. Ancestry outliers were removed so that only European samples were kept using principal component analysis (PCA) with a 1000 Genomes reference (Hansen, et al., 2019). Markers with >0.01 missingness, <0.01 minor allele frequency (MAF) and/or Hardy-Weinberg equilibrium P <0.0001 were removed. From this dataset, we extracted the DLD cases and controls. We performed the following study-specific QC: we checked for family relatedness in the remaining individuals with PLINK v1.90b6.21 following a published protocol (Anderson et al., 2010), and, using a script from the same publication, one individual in each pair of related individuals was removed based on missingness. Principal components (PCs) were computed in the remaining individuals in the genotyped dataset with PLINK, to be used as covariates.

Imputation of further markers
A joint variant calling with Graphtyper (Eggertsson et al., 2017) forms the basis of the imputation. Imputation was done using the in-house workflow developed by deCODE (Gudbjartsson et al., 2015).

Comparison with medical registers
The diagnostic information obtained from the hospital registers showed that only 19 individuals among all DBDS individuals for whom we had questionnaire data had received a (primary or secondary) diagnosis of language disorder, either ICD-10 code F80.1 or F80.2. Interestingly, these codes were sometimes given to an individual on the same day, even though ICD-10 guidelines specify they are mutually exclusive. Thus, the prevalence of diagnoses of language disorder from the registers was 0.04%, compared to the prevalence of DLD in our sample (3.36% in DBDS; 3.70% among DLD cases+controls). Interestingly, only one person with either of these ICD-10 diagnoses was a DLD case, and 4 DLD controls had one of those diagnoses as well (the rest of these were not included as cases or controls). school language unit (green), which are likely to be more intensive, do not seem to differ.

DLD interventions over the years
Over time, these two types of intervention together with other form of language intervention (orange) were reported markedly less frequently than speech-language therapy.

Comorbidities between DLD and reading impairment or learning difficulties
Results for the questions on the two main comorbidities are shown in

Effect of DLD status on physical and mental health scores
We observed significant effects of DLD status on both health component summary scores from the SF-12 questionnaire. The presence of DLD was associated with a reduced mental health component summary score of β=-0.912 (95% CI: -1.286-(-0.538), P=1.77×10 -6 ) as well as a reduced physical component summary score of β=-0.557 (95% CI: -0.804-(-0.309), P=1.05×10 -5 ). Supplementary Figure S1 shows the results of all regressions.

Genetic analyses
We found one association locus reaching genome-wide significance (Figure 3

Low prevalence of DLD in a healthy adult population
The prevalence of DLD in this study was lower than previous reports in school-aged children (Norbury, et al., 2016;Tomblin, et al., 1997), at 3.70 % among cases+controls.
Identification of DLD cases was based on retrospective self-reported language difficulties and having received language intervention during childhood, not on prior or current assessment of language impairment symptoms. Thus, we expect that recall bias is likely to contribute to the low prevalence, especially given that not all children with DLD receive intervention (Tomblin, et al., 1997). Also, we would miss cases who self-reported language difficulties but did not receive language intervention or vice versa, but it is unlikely that these would be misclassified as DLD controls due to the DLD control inclusion criteria. The prevalence of DLD was much lower (0.04%) when considering ICD diagnoses from the hospital registers. To our knowledge, there are no published studies which formally assessed DLD prevalence in adult populations. Therefore, there is a possibility that the prevalence in an older sample would be lower, but it is not likely to decrease from ~7% to 0.04%; studies have shown little evidence to suggest a decline in prevalence until at least age 16 (Law, Boyle, Harris, Harkness, & Nye, 2000). In another DBDS study, a discrepancy between formal diagnosis and reported symptoms was observed as well. The study showed that reported attention deficit/hyperactivity disorder (ADHD) symptoms are relatively common even among undiagnosed individuals (Hoeffding et al., 2018). Furthermore, it has been shown that individuals in the DBDS sample have a higher education level than the general population (Burgdorf et al., 2017), which probably contributed to the lower DLD prevalence (3.70%) in this sample. Given that DLD and educational attainment are negatively correlated there are probably fewer individuals with DLD in the DBDS sample than in the Danish adult population. Lastly, the "healthy donor effect" may also be at play here. The "healthy donor effect" pertains to donors being a selected "healthier" subset of a population due to both donor selection procedures and self-selection (van den Hurk, Zalpuri, Prinsze, Merz, & de Kort, 2017).
Our study found an association between DLD and the male sex. This result is in accordance with a previous study of case history factors (Rudolph, 2017). However, several population-based studies did not find significant sex differences (Calder, Brennan-Jones, Robinson, Whitehouse, & Hill, 2022;Norbury, et al., 2016;Tomblin, et al., 1997). It thus seems likely that referral bias contributes to the association between DLD and the male sex (Evans, 2010). This suggests that females may be less likely to receive language intervention.
In the hospital registers, very few individuals from DBDS received an ICD diagnosis of language disorder. The diagnoses recorded in the Danish hospital registers are rarely given by personnel with expertise in speech and language pathology as part of the diagnostic team. This leads to a lack of awareness of DLD in these systems, as reflected in the extremely low prevalence of 0.04% (compared to 3.70% in DLD cases+controls, based on the questionnaire). Furthermore, these registers reflect diagnoses given through hospital contact only, which may not apply to most individuals seeking identification of their language disorders. The discrepancy in prevalence indicates that most individuals with DLD go undiagnosed in the hospital registers, in contrast to e.g. autism spectrum disorder . Among the 5 individuals who overlapped between the DLD case-control sample and the hospital registers with F80.1 or F80.2, 1 was a DLD case and 4 were DLD controls. If these 4 DLD controls are false negatives in our classification, it could be due to their not remembering, not knowing or not having been told that they had language difficulties (recall bias). Alternatively, it could be that the language disorder diagnoses from the hospital registers are not reliable (some individuals received mutually exclusive diagnoses on the same day). Lastly, the lower DLD prevalence in this study suggests that neither the educational system nor the hospital system identify all children with DLD. This calls for national guidelines for DLD to secure both identification and services. One limitation of this study, however, is that the identification of cases and controls is based exclusively on retrospective self-report, which likely contributed to the low prevalence. Future studies should also include standardized language assessment for phenotyping.

No trends of change over time in types of DLD intervention
According to self-report, the large majority of the DLD cases received intervention of the type speech-language therapy, which likely comprised individual intervention given by a speech-language therapist. No marked trends of change in intervention types reported according to birth year were detected in the data. Thus, the suggested decrease in special education (de López & Knudsen, 2019) is not confirmed in the present study. This is not surprising, given that our sample comprises adults. A limitation in the present study is that we did not ask specifically about experience of consultation-based intervention. The reason for that was that we did not expect participants to have reliable recollection of such childhood services. If some individuals did remember receiving consultation-based intervention, they probably responded "yes" to question 3d about other form of language intervention (Supplementary Material). As indicated in Figure 2, there was no clear change over time for this form of intervention. The expected increase over time of inclusive education was not confirmed by the present study (de López & Knudsen, 2019).

Higher prevalence of reading and learning difficulties in DLD
DLD status was highly associated with both reading and learning difficulties. The prevalence of reading impairment in DLD cases was ~19%. This represents the lower range of previously reported prevalence in population-based DLD samples of schoolaged children (Catts, Adlof, Hogan, & Weismer, 2005;Nippold & Tomblin, 2014). The prevalence of reading impairment in controls, ~6%, aligns well with the estimated prevalence in the general populations (Elbro, et al., 2011;Hulme & Snowling, 2011). The relatively low prevalence of reading and learning difficulties in the present study could be a consequence of using written questionnaires, since cases with DLD may have opted out of answering the DBDS questionnaire due to reading difficulties. In a long-term followup on reading in adults with DLD, Botting (Botting, 2020) found that reading questionnaires was particularly difficult: 48.1% of the DLD sample found this type of reading difficulties, as compared with 4.7% of controls. In future studies, participants could be provided with the option of assisted response to the questionnaire by, for example, having the questions read aloud and explained (Botting, 2020); supplementary Table S2 indeed shows that a considerable proportion (at least 43.55%) of the DLD cases received some form of reading intervention.

Lower self-reported health outcomes in DLD
Mental and physical health was negatively associated with DLD. The effect of having an indication of DLD was a reduction of approximately one point in the mental health component summary score and half a point in the physical component summary score.
The clinical significance of these reductions has not been investigated; to put these effects in context, we compared them to the effects of inflammation on the SF-12 composite scores. For physical health, the effect of DLD (β=-0.557) is about one third of the effect of inflammation (β=-1.53), as previously reported for linear regression models in DBDS (Dinh, et al., 2019). For mental health, the effect of DLD (β=-0.912) is about 7 times larger than the effect of inflammation (β=-0.13), although the latter was not statistically significant in that study.
Poor mental health in adults with DLD has been reported in a previous study (Botting, et al., 2016), albeit not one which used a Scandinavian cohort. Our results illustrate the need for raising awareness of the link between language disorder and mental health in medical and psychiatric services. In addition, knowledge of language disorders, and how communication may be affected by them, needs to be present at all levels of medical services in order to provide individuals with language disorders with appropriate healthcare. Currently, there is a risk of disadvantage for individuals with DLD in getting attention and support for health-related issues (Botting, et al., 2016;Durkin & Conti-Ramsden, 2010).

DLD does not show non-zero heritability or overlap with main SLI genetic loci
If we take the results of our genome-wide analyses at face value, DLD status was not heritable in our sample, in contrast to assessments for SLI. Here we estimated the narrowsense SNP heritability, which is influenced by the number of cases and controls. Thus, it could be that the number of DLD cases with genotype data (N=610) was the limiting factor. Also, it is likely that the generalized nature of our questions contributed to DLD case heterogeneity (and hence to DLD case genetic heterogeneity). This could result in a lower heritability estimate, since genetic variance (and therefore increased heritability) results from genotypes of cases being more similar to genotypes of other cases, and genotypes of controls being more similar to genotypes of other controls (S. H. Lee et al., 2013). However, the same genetic heterogeneity among DLD cases could be an outcome of the broader scope of DLD compared to SLI. Indeed, for SLI, recent analyses using SLI-trained PRS showed a significant adjusted R 2 of >5% for SLI in an independent target sample (Nudel, et al., 2021;Nudel, et al., 2020); when the target phenotype was broader and included children with low language scores irrespective of their non-verbal intelligence or indication of autism spectrum disorder, the predictive ability of the SLItrained PRS dropped considerably (Nudel, et al., 2020). In the same vein, our result suggest that DLD, as a clinical diagnosis, is a "mixed bag" when it comes to molecular etiology, although the number of DLD cases in the GWAS is likely to have been a contributing factor to this result. In this regard, it is also important to consider that our phenotype was based on questionnaire data and not on clinical assessment or on standardized tests, as often employed in SLI studies, and that it did not measure the severity of the language deficits. An analogy for the DLD-SLI heritability discrepancy could be that of the difference between minimally phenotyped major depressive disorder (MDD) and fully diagnosed MDD; it has recently been shown that minimal phenotyping for MDD (i.e. based on self-report, questionnaires about symptoms filled out by the participant or help seeking) results in a large, non-specific genetic component and lower SNP heritability estimates, which could not be attributed to the inclusion of "milder" MDD cases (Cai et al., 2020). Additionally, some of the GWAS hits identified with 23 minimal phenotyping of MDD were not specific to MDD. If we are observing a similar phenomenon here, it could imply that, while DLD criteria might be useful in the clinical setting, they could be less useful for studying the molecular pathways that underlie developmental pathologies of language, which requires more specific or narrowly defined phenotypes.
Despite the fact that the overall h 2 SNP in our study did not significantly differ from zero, we identified one genome-wide significant locus, on chromosome 8p12. The lead SNP, rs183826546, is located within a non-coding RNA gene, LOC105379365. Interestingly, chromosome 8p12 was a suggestive linkage region in the first SLI Consortium study (The SLI Consortium, 2002). Furthermore, chromosome 8p in general and 8p12 in particular have been implicated in many neurodevelopmental conditions (Tabares-Seisdedos & Rubenstein, 2009), and this result might therefore represent an association with general or non-specific neurodevelopmental deficits, which are captured by our DLD phenotype. To the best of our knowledge, this is the first reported genome-wide significant association for DLD in a case-control GWAS.

CONCLUSIONS
Our study is the largest DLD study incorporating both population-level epidemiological and genetic analyses. In Denmark, where speech-language services are separate from the healthcare system, DLD is critically underdiagnosed in the hospital registers, which could be detrimental, considering that it was associated with negative physical and mental health outcomes in our study, as well as with reading impairment and learning difficulties. Since we included secondary diagnoses as well, when examining hospital registers, this means that a language disorder is typically not given even when the patient might be in contact with the hospital due to associated health problems. Our genetic analyses found stark differences between DLD and previous reports for SLI, whereby we did not observe significant non-zero heritability for DLD. If this is not the result of the small number of cases, this could suggest that the DLD diagnosis might be too broad for molecular genetic studies, which is supported by the fact that the only genome-wide significant locus identified in our study overlapped with loci for other neurodevelopmental disorders. We believe that raising awareness of DLD in the healthcare system and having a more integrative approach towards DLD diagnosis and treatment could greatly benefit affected individuals.

Data availability
We encourage scientific collaborations using the data generated in the Danish Blood Donor Study (DBDS) Genetic Consortium. Published summarized data are available on request. Otherwise request of data necessitates first approval by the DBDS steering committee and if the request is considered outside the aim of DBDS, application to the national scientific ethical committee is obligatory. Additionally, material transfer and data protection agreement need to be acquired. Please visit http://www.dbds.dk.

Acknowledgements
Donor Study as well as the staff at the blood centers making this study possible.